flutter_kitten_tts

High-quality offline text-to-speech for Flutter using the KittenML v0.8 ONNX model.

Features

• Fully offline – model runs on-device, no internet after initial download
• 8 voices – Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, Leo
• Adjustable speed – 0.5x to 2.0x
• Automatic chunking – long texts are split at sentence boundaries
• 24 kHz output – PCM Float32 audio ready for playback

Architecture

Text → TextPreprocessor → espeak-ng (FFI) → TextCleaner → ONNX Runtime → PCM audio
         (numbers,          (IPA            (token         (inference)
          currency,          phonemes)       mapping)
          time, etc.)

Setup

1. Add dependency

From pub.dev (recommended):

dependencies:
  flutter_kitten_tts: ^0.0.4

From Git (latest):

dependencies:
  flutter_kitten_tts:
    git:
      url: https://github.com/ikeoffiah/kitten_tts_flutter.git
      ref: main

Local path (development):

dependencies:
  flutter_kitten_tts:
    path: ../kitten_tts_flutter

2. espeak-ng source (Android)

Android builds espeak-ng from source. The package (both the GitHub repo and the pub.dev publish) includes third_party/espeak-ng, so you do not need to run any script—add the dependency and build. iOS/macOS use bundled sources in the plugin and also require no extra steps.

If you see Android build errors such as "espeak-ng source not found" or missing espeak-ng files, run the setup script once from the plugin directory (e.g. from your pub cache after flutter pub get):
bash "$(flutter pub cache path)/hosted/pub.dev/flutter_kitten_tts-0.0.4/scripts/setup_espeak.sh"
(Replace 0.0.4 with your installed version, or for a git dependency use .../git/kitten_tts_flutter-*/scripts/setup_espeak.sh.) This downloads espeak-ng 1.52.0 into third_party/espeak-ng/.

3. Platform configuration

iOS/macOS: espeak-ng is compiled from source via CocoaPods automatically.

Android: espeak-ng is compiled via CMake (configured in build.gradle).

Usage

import 'package:flutter_kitten_tts/flutter_kitten_tts.dart';

final tts = KittenTTS();

// Initialize (downloads ~35 MB model on first run)
await tts.initialize(
  onProgress: (progress, status) {
    print('$status (${(progress * 100).toStringAsFixed(0)}%)');
  },
);

// Generate speech
final audio = await tts.generate(
  'Hello world! This is KittenTTS.',
  voice: 'Jasper',  // Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, Leo
  speed: 1.0,       // 0.5 to 2.0
);

// audio is Float32List at 24000 Hz – play it with audioplayers, just_audio, etc.
print('Generated ${audio.length} samples (${audio.length / 24000}s)');

Streaming (chunked generation)

For long texts, generate chunks individually for streaming playback:

final chunks = tts.splitText(longText);
for (final chunk in chunks) {
  final audio = await tts.generateChunk(chunk, voice: 'Luna');
  // Play each chunk as it's ready
}

Available voices

Voice	Gender	Key
Bella	Female	expr-voice-2-f
Jasper	Male	expr-voice-2-m
Luna	Female	expr-voice-3-f
Bruno	Male	expr-voice-3-m
Rosie	Female	expr-voice-4-f
Hugo	Male	expr-voice-4-m
Kiki	Female	expr-voice-5-f
Leo	Male	expr-voice-5-m

Model files

On first initialization, the package downloads from HuggingFace:

File	Size	Description
kitten_tts_nano_v0_8.onnx	~24 MB	ONNX model (int8)
voices.npz	~3 MB	Voice embeddings
espeak-ng-data.tar.bz2	~7 MB	Phonemization data

Files are cached in the app's support directory and persist across app launches.

Requirements

• Flutter >= 3.3.0
• Dart >= 3.10.8
• iOS 13.0+ / macOS 10.14+ / Android API 24+

Contributing

If you clone this repo (directory may be named kitten_tts_flutter) and see analysis errors like "Target of URI doesn't exist: package:flutter_kitten_tts/...", run after flutter pub get:

bash scripts/fix_package_config.sh

Then run dart analyze or flutter test as usual.

License

MIT